IGBT Remaining Useful Life Prediction Based on SVMD-WGAN-BiLSTM

A novel approach is proposed to address the issues of insufficient generalization capability and low prediction accuracy in existing lifetime prediction methods for insulated gate bipolar transistors(IGBT). A bidirectional long short-term memory(BiLSTM) model with successive variational mode decomposition(SVMD) and a Wasserstein distance-based adversarial generative neural network(WGAN) are integrated in the approach. Firstly, the original sequence is decomposed using SVMD to obtain multiple intrinsic mode components and residual components. Subsequently, components with high Pearson correlation coefficients are selected and superimposed to construct a denoised new sequence. This new sequence is then subjected to data augmentation via the WGAN, a robust adversarial network leveraging the Wasserstein distance for stability and quality of generated samples. Finally, the newly constructed sequences are ultimately input into a bidirectional LSTM neural network for remaining useful life prediction and are validated using NASA's IGBT accelerated aging test data. The results indicate that the model achieves an average absolute error of 0.707%, a mean squared error of 0.013 2, and a fitness degree of 0.974. Compared with BiLSTM, SVMD-BiLSTM, GRU, and CNN algorithms, the prediction accuracy of the SVMD-WGAN-BiLSTM model has improved significantly, demonstrating excellent generalization performance and providing a valuable reference for IGBT remaining useful life prediction.